28 de setembro de 2020 - 7h58

Não seria totalmente inesperado, nem surpreendente, mas não deixa de ser intrigante: a China está usando Inteligência Artificial para gerar rostos falsos e engajar a audiência do Facebook em ações de propaganda política nos EUA, envolvendo campanhas pró e contra Trump e seus adversários políticos mais diretos.

A descoberta é do instituto de análises tecnológicas contra conteúdos fake na web Graphika. Como relata o AdAge: “Num sinal dos tempos, Graphika descobriu que a rede chinesa de contas falsas emprega rostos criados por meio de uma técnica de IA conhecida como GANs (Generative Adversarial Networks). Esses rostos falsos são utilizados para iludir a detecção, mas como sua assinatura visual geralmente termina com peculiaridades e anomalias sutis, às vezes os GANs podem ser facilmente detectados. “Esta forma de IA está prontamente disponível online e seu uso (ou abuso) por operações secretas explodiu no ano passado”, de acordo com o relatório do Graphika, que identificou uma dúzia de imagens geradas pelo GAN da operação de informação chinesa.

Para buscar eliminar ou minimizar os efeitos dessa atividade política de influenciamento através de conteúdos falsos, o Facebook retirou do ar 155 contas, 11 páginas, 9 grupos e 7 contas do Instagram conectadas a atividades de operações vindas da China em direção aos EUA.

Mas identificou também as mesmas técnicas chinesas para influenciar a situação política nas Filipinas. Veja abaixo texto da reportagem do Adage.

Facebook  removed two networks of fake accounts spreading government propaganda on the platform Tuesday, one originating in China and one in the Philippines.

In its latest report on this kind of coordinated campaign, the company says it took down 155 Facebook accounts, 11 pages, nine groups and seven Instagram accounts connected to the Chinese activity and 57 accounts, 31 Pages and 20 Instagram accounts for the activity in the Philippines. Both operations broke Facebook’s rules against “coordinated inauthentic behavior on behalf of a foreign or government entity.”

The company released the report Thursday in coordination with Graphika, a social analytics company that specializes in disinformation. Graphika  regularly analyzes this kind of activity in coordination with Facebook and its reports dive into more depth about techniques.

In a sign of the times, Graphika found that the Chinese network of fake accounts employed faces created through an AI technique known as GANs (Generative Adversarial Networks). Those fake faces are employed to elude detection, but because their visual signature often ends up with subtle quirks and anomalies, GANs can sometimes be easily detected. “This form of AI is readily available online, and its use (or abuse) by covert operations has exploded in the last year,” according to Graphika’s report, which identified a dozen GAN-generated images from the Chinese information operation.

“A year ago, this was a novelty,” Graphika’s Ben Nimmo wrote on Twitter. “Now it feels like every operation we analyse tries this at least once.”

GANs examples via Graphika

The Chinese campaign, which Facebook traced to China’s Fujian province, included a “small volume” of activity directed at the U.S. election. Those efforts, which began in April 2019 with a Facebook group called Go for Pete Buttigieg 2020, did not gain much traction. That account never added members beyond two accounts affiliated with the propaganda network. In 2020, the network of fake accounts created three U.S.-focused groups, one pro-Trump, one pro-Biden-Harris and one called “Quack Quack” that made anti-Trump posts. The Biden-Harris group found the most success, attracting 1,400 members.

Most of the newly identified Chinese propaganda pushed China’s interests in the Philippines and Southeast Asia. Unlike some campaigns, the people involved in this activity took steps to hide their identities using VPNs and other tactics. According to Graphika, the network of fake accounts “showed a particular interest in maritime security, especially in the South China Sea.” This is the second set of Facebook takedowns for Chinese information operations, following a previous report in August 2019.

“In Southeast Asia where this network focused most of its activity, they posted in Chinese, Filipino and English about global news and current events including China’s interests in the South China Sea; Hong Kong; content supportive of President Rodrigo Duterte and Sarah Duterte’s potential run in the 2022 Presidential election; criticism of Rappler, an independent news organization in the Philippines; issues relevant to the overseas Filipino workers; and praise and some criticism of China,” Facebook wrote in a blog post on the takedowns.

The other propaganda campaign originated in the Philippines, and an investigation identified ties to the country’s national military and police. Two pages in the campaign boasted more than 40,000 followers but others had none or few. Those accounts shared content in Filipino and English about “local news and events including domestic politics, military activities against terrorism, [a] pending anti-terrorism bill, criticism of communism, youth activists and opposition, the Communist Party of the Philippines and its military wing the New People’s Army, and the National Democratic Front of the Philippines,” according to Facebook.

(*) Com material do AdAge